How Do I Get ChatGPT to Cite My Website? What We Measured

First, ChatGPT's search crawler has to read you: in 15 days it read 328 of our ~1,770 URLs and none of our AI service pages. Then, answer specific questions with your own data and be present in the third-party sources ChatGPT cites. Here is our data, not theory.

See our GEO service See the monthly measurement
  • ~7 a day Visits we get from ChatGPT
Scroll

Definition

How do you get ChatGPT to cite your website?

For ChatGPT to cite your website, its search crawler (OAI-SearchBot) must be able to read it, and the page must answer exactly what the user asked; otherwise ChatGPT will cite third parties that talk about you. On kiwop.com that crawler read 328 of ~1,770 URLs in 15 days, and the ~7 daily visits we get from ChatGPT land on pages that answer specific questions, not on "best agency" pages.

Everything below comes from two sources of our own: kiwop.com's server logs from 11 to 25 September 2026, with each crawler's IP checked against the lists OpenAI publishes, and the monthly measurement by Kiwop Labs, which asks ChatGPT 40 buying questions and publishes the full answers. We keep what we measured apart from what is a hypothesis: a page being read by the crawler looks like a precondition for being cited, but that is a correlation we observe, not a proven cause.

In numbers

What We Measured on kiwop.com

nginx logs from 11 to 25 September 2026 and the September GEO baseline.

  • 328 / ~1,770 URLs read by OAI-SearchBot In 15 days, at 30-60 URLs a day. Fewer than one in five sitemap pages.
  • 26 of 121 Main Spanish-language pages The only ones it read. None of our AI service pages.
  • 106 Visits from ChatGPT in 15 days About 7 a day (visit = IP × day), 91% on desktop.
  • 2 of 40 "Which agency" questions that mention us ChatGPT (gpt-5.2 API with web search), September 2026 measurement.

Key points

The 6 Things Our Data Says Matter

In order: without the first, the rest will not do much.

  1. 01

    Get read by the right crawler

    OpenAI runs three crawlers. OAI-SearchBot feeds ChatGPT search, and OpenAI says sites that block it are not shown in its search answers. GPTBot is for model training and ChatGPT-User acts when a user asks it to. Check robots.txt, your firewall and your CDN (Cloudflare can block AI bots), and make sure your server does not throw errors: we had zero 5xx errors in the period.

  2. 02

    Nudge it towards the pages it skips

    If the crawler only reads 30-60 URLs a day, it chooses which. Since 25 September 2026 we send a nightly IndexNow ping (Bing and other search engines) for the pages OAI-SearchBot has not read in 30 days, alongside a clean sitemap and Bing Webmaster Tools. We do not know the result yet: we will measure it in October.

  3. 03

    Answer one specific question with your own data

    Our visits from ChatGPT land on pages that settle a precise question: GEO in English, marketing attribution in French, headless ecommerce in German, MCP protocols or the cost of ISO 27001. They do not come from "best agency". One question per page, the answer in the first sentence, every figure with a source.

  4. 04

    Be in the third-party sources it cites

    Across 40 "which agency do you recommend?" questions, ChatGPT cited 117 different domains and none more than 3 times: clutch.co, sortlist.es and c6n.eu (3), europapress.es, elpais.com and cincodias.elpais.com (2), plus the agencies' own websites. No single source decides it: you need to be in several (review directories, press, rankings with a published method).

  5. 05

    Measure with data that doesn't lie

    In 15 days we received ~1,100 requests carrying an OpenAI crawler name from IPs that are not OpenAI's. If you do not verify the IP, you count fake bots. People arriving from ChatGPT are measured by the utm_source=chatgpt.com link. How we do it: how to measure traffic from ChatGPT.

  6. 06

    Don't waste time on llms.txt

    Our llms.txt got 93 requests in 15 days and none from the OpenAI, Anthropic, Google or Perplexity crawlers; just one from Common Crawl. Today it is not what gets you cited by ChatGPT. Full data in does llms.txt do anything?.

Who it is for

Is This Approach for You?

It works better for some websites than others. Better to know upfront.

Who it's for

  • Companies whose buyers research before purchasing (B2B, services, tech) and ask an assistant
  • Websites that can publish their own data: prices, measurements, cases, real experience
  • Teams with access to server logs, or who can get them from their host
  • Anyone willing to measure for months before drawing conclusions

Who it's not for

  • Anyone who wants to show up tomorrow for "the best X company": we appear in 2 of 40 questions like that
  • Websites with nothing of their own to say, repeating what a thousand others already say
  • Anyone looking for a guarantee that ChatGPT will cite them: nobody can give one
  • Websites that cannot touch robots.txt, the firewall or the CDN

FAQ

FAQs About Getting Cited by ChatGPT

Answered with our data. Where we do not know, we say so.

Which OpenAI crawler do I need to allow to show up in ChatGPT?

OAI-SearchBot. According to OpenAI, it is the one used to surface websites in ChatGPT search, it respects robots.txt and, if you block it, your site will not appear in its search answers. A robots.txt change takes about 24 hours to apply. GPTBot is for model training (you can block it without losing search) and ChatGPT-User acts when a user asks it to, without following robots.txt.

How many pages of my website does ChatGPT read?

It depends on the site; your logs are the only way to know. On ours, OAI-SearchBot read 30-60 URLs a day: 328 distinct URLs in 15 days, out of about 1,770 in the sitemap. Of our 121 main Spanish-language pages it read 26, and no AI service pages. We do not know how it chooses; that is why we notify it of the pages it skips.

How do I know a visit really comes from the ChatGPT crawler?

Check the IP against the lists OpenAI publishes: searchbot.json, gptbot.json and chatgpt-user.json on openai.com. The user-agent is not enough: in 15 days we received ~1,100 spoofed requests using OpenAI's name from other IPs (394 as ChatGPT-User, 382 as OAI-SearchBot and 349 as GPTBot).

Does llms.txt help you appear in ChatGPT?

On our data, no. In 15 days our llms.txt received 93 requests: none from the OpenAI, Anthropic, Google or Perplexity crawlers, and only one from Common Crawl. We go into detail in does llms.txt do anything?.

Which sources does ChatGPT cite when you ask it about suppliers?

In our September 2026 measurement (40 "which agency do you recommend?" questions, gpt-5.2 API with web search), ChatGPT gave sources in 21 answers and cited 117 different domains, none more than 3 times. The most repeated: clutch.co, sortlist.es and c6n.eu (3 answers each); europapress.es, elpais.com and cincodias.elpais.com (2); and many agency websites. The full answers are in the GEO baseline.

How long does ChatGPT take to cite a new page?

We do not know for sure, and be wary of anyone who gives you a fixed timeline. The only thing OpenAI publishes is that a robots.txt change takes about 24 hours to apply. Since 25 September 2026 we send a nightly IndexNow ping for the pages its crawler has not read in 30 days; in October we will publish whether that changes anything.

How much traffic does ChatGPT send to a website?

To ours, 106 visits in 15 days (about 7 a day, counting one visit per IP per day), 91% on desktop. The volume is small, but it arrives with a specific question. To measure it properly without inflating it, see how to measure traffic from ChatGPT.

How much does it cost to have an agency work on ChatGPT visibility?

Kiwop's GEO service starts at €1,500/month (published price; finalised at kickoff depending on markets, languages and sector). Before paying anyone, check the free things: that robots.txt and your CDN let OAI-SearchBot in, that your server does not throw errors, and which pages it actually reads in your logs.

Sources

Sources for the Figures on This Page

Our own data and public documentation. Accessed 25 September 2026.

  1. kiwop.com nginx logs, 11-25 September 2026 (our own data) URLs read by OAI-SearchBot (328 of ~1,770 in the sitemap; 26 of 121 main Spanish-language pages; zero 5xx), ~1,100 requests with an OpenAI user-agent from non-OpenAI IPs, 106 visits with utm_source=chatgpt.com (visit = IP × day; 91% desktop) and 93 requests to /llms.txt. IPs verified against OpenAI's lists.
  2. Kiwop Labs, GEO baseline dataset (September 2026) Generated on 5 Sep 2026. Provider "openai" (gpt-5.2 with web search), 40 questions: Kiwop mentioned in 2; 21 answers with sources and 117 domains cited (cited_domains field).
  3. Kiwop Labs, GEO baseline: method and full answers How the questions are run and what is extracted from each answer.
  4. OpenAI, Overview of OpenAI Crawlers What OAI-SearchBot, GPTBot and ChatGPT-User do, whether they respect robots.txt, and the ~24 hours a change takes to apply.
  5. OpenAI, OAI-SearchBot IP list Used to verify the requests in our logs, together with gptbot.json and chatgpt-user.json.
  6. IndexNow Protocol for notifying search engines of changes; used by Microsoft Bing, Naver, Seznam.cz, Yandex and Yep.
  7. kiwop.com robots.txt Open to all crawlers, AI crawlers included.

Next step

See What ChatGPT Reads on Your Site Before Investing in Content

We review it with you: which pages its crawler reads, which third-party sources it cites in your sector and what is missing for it to cite you. If a few technical fixes are enough, we will tell you.

Talk about my case
  • No commitment
  • Response in 24h
  • Custom proposal
Last updated: September 2026

Let's talk.

Initial technical consultation

AI, security and performance. Diagnosis with phased proposal.

  • NDA available
  • Response <24h
  • Phased proposal

Your first meeting is with a Solutions Architect, not a salesperson.

Request diagnosis